Struct ReproducibilityManager 

pub struct ReproducibilityManager {
    pub environments: HashMap<String, EnvironmentSnapshot>,
    pub reports: Vec<ReproducibilityReport>,
    pub verifications: Vec<VerificationResult>,
    pub config: ReproducibilityConfig,
}

Reproducibility manager for tracking experiment reproducibility

Fields

environments: HashMap<String, EnvironmentSnapshot>

Environment snapshots

reports: Vec<ReproducibilityReport>

Reproducibility reports

verifications: Vec<VerificationResult>

Verification results

config: ReproducibilityConfig

Configuration

Implementations

impl ReproducibilityManager

pub fn new(config: ReproducibilityConfig) -> Self

Create a new reproducibility manager

pub fn capture_environment(&mut self, random_seeds: &[u64]) -> Result<String>

Capture current environment snapshot.

random_seeds should be the actual seed(s) the experiment being snapshotted used; pass an empty slice if none were recorded. This used to be hardcoded to vec![42] regardless of what the experiment actually did, which made the “random seed documented” checklist item trivially true for every snapshot rather than reflecting reality.

pub fn generate_report( &mut self, experiment_id: &str, environment_id: &str, ) -> Result<String>

Generate reproducibility report for experiment

pub fn verify_reproducibility( &mut self, original_experiment_id: &str, reproduction_experiment_id: &str, original_metrics: &HashMap<String, f64>, reproduction_metrics: &HashMap<String, f64>, original_environment_id: Option<&str>, reproduction_environment_id: Option<&str>, ) -> Result<String>

Verify reproducibility between two experiment runs by comparing their actual final metrics.

Every key present in either original_metrics or reproduction_metrics is compared (relative to config.numerical_tolerance); a key named with a performance-ish marker (time/memory/latency/throughput/duration/speed) is scored as performance_similarity, everything else as result_similarity, so wall-clock jitter between runs cannot mask (or be masked by) an actual difference in the computed result, or vice versa.

When original_environment_id/reproduction_environment_id are both supplied and resolve to a captured EnvironmentSnapshot, environment_similarity is computed for real from OS/architecture and pinned-dependency-set overlap; otherwise it is None rather than a fabricated number. configuration_similarity is always None: this function has no experiment-configuration data to compare against.

Trait Implementations

impl Clone for ReproducibilityManager

fn clone(&self) -> ReproducibilityManager

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ReproducibilityManager

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for ReproducibilityManager

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for ReproducibilityManager

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.